|
It looks like Dalle-E-2 has less of an idea of what Frank Zappa, or a sci-fi cover illustration, actually looks like. But actual lol at both the name and pic of Fapank Z Fankick.
|
#
?
Jul 4, 2022 21:27
|
|
|
|
| # ? May 30, 2024 13:56 |
|
|
The words that these things spit out are some of my favorite. They're kind of like the stuff mad magazine would come up with for parody names. I like Frank Fatank!
|
|
#
?
Jul 4, 2022 21:33
|
|
|
e, I need to retool my prompt.
LifeSunDeath fucked around with this message at 21:42 on Jul 4, 2022 |
|
#
?
Jul 4, 2022 21:39
|
|
|
Here's midjourney attempts at the frank zappa and magician prompts  And here's Hiroshi Nagai doing the skyline of Ancient Rome 
|
|
#
?
Jul 4, 2022 22:00
|
|
|
axolotl farmer posted:It looks like Dalle-E-2 has less of an idea of what Frank Zappa, or a sci-fi cover illustration, actually looks like. I added a couple words to get more of that old sci fi magazine feel.  "frank zappa sci-fi pulp paperback cover colorful illustration 1960's"  “frank zappa driving a moon rover on a alien planet sci-fi pulp paperback cover colorful illustration 1960's”
|
|
#
?
Jul 4, 2022 22:32
|
|
|
axolotl farmer posted:It looks like Dalle-E-2 has less of an idea of what Frank Zappa, or a sci-fi cover illustration, actually looks like. It's good. They all have that amazing stache which was his signature. And also, it's a cartoonist sci-fi cover so detail is not important.
|
|
#
?
Jul 4, 2022 23:20
|
|
|
Now these are definitely more disturbing. That top left one is especially great. I've noticed that some of the image generators it treats the words in prompts as weights, so if you put the same word multiple times it'd weigh it more heavily for the final image. Like, if I put "a grinning grinning grinning man" it would generate images of men with grins extending off the sides of their face and/or rising up past their eyeballs, usually with multiple additional rows of teeth. Do DALL-E-2 and Midjourney work the same way?
|
|
#
?
Jul 5, 2022 03:47
|
|
|
Pitdragon posted:Now these are definitely more disturbing. That top left one is especially great. which ones do the thing you mention? I've noticed it having a mild effect in mini dall-e/craiyon, but more of just weighting in a certain direction and not really the multiplicative effect you're talking about like if I put in "yoda chewbacca yoda" it would give me some output that was closer to yoda than chewbacca I tried your grinning example on craiyon and got pretty similar results for "a grinning man" as I did for "a grinning grinning grinning man"   and I tried replicate too 
|
|
#
?
Jul 5, 2022 04:05
|
|
|
TIP posted:which ones do the thing you mention? I've noticed it having a mild effect in mini dall-e/craiyon, but more of just weighting in a certain direction and not really the multiplicative effect you're talking about I've been using https://replicate.com/jack000/glid-3-xl for these, with the guidance scale at 17.77 and the seed at -1, as you said the effect isn't really as strong in mini dall-e Here's "a grinning man"  Okay, screwed up the arms, but the grin isn't too awful. "A grinning grinning grinning man"  Now we've got extra rows of teeth and the mouth is extra long. Also the eyes are getting a little crazy. "A grinning grinning grinning grinning grinning grinning man"  Still getting extra rows of teeth and the huge cartoonish mouth, but now it seems to also always try to make the eyes into mouth or teeth like things.
|
|
#
?
Jul 5, 2022 04:56
|
|
|
EVIL Gibson posted:It's good. They all have that amazing stache which was his signature. They got the mustache, but all of them miss the soul patch.
|
|
#
?
Jul 5, 2022 05:06
|
|
|
Could someone with a better model than Dall-e try "Lawyer dog making an opening statement in front of a jury"? It puts together the concepts perfectly but the execution
|
|
#
?
Jul 6, 2022 01:14
|
|
|
Well this technique of shrinking and reuploading to uncrop is pretty drat cool https://www.reddit.com/r/dalle2/comments/vr3wfm/painting_of_a_cinematic_still_of_a_female/ Start with this:  Uncrop to this (took a bunch of tries to get something usable):  Uncrop again:  Start with this:  Uncrop directly to this: 
|
|
#
?
Jul 6, 2022 01:19
|
|
|
drat I want the uncrops to keep going and going
|
|
#
?
Jul 6, 2022 01:50
|
|
|
Snowy posted:drat I want the uncrops to keep going and going they say if you uncrop enough you see your own death.
|
|
#
?
Jul 6, 2022 02:04
|
|
|
LifeSunDeath posted:they say if you uncrop enough you see your own death. 
|
|
#
?
Jul 6, 2022 02:05
|
|
|
Fintilgin posted:Well this technique of shrinking and reuploading to uncrop is pretty drat cool https://www.reddit.com/r/dalle2/comments/vr3wfm/painting_of_a_cinematic_still_of_a_female/ Uncrop likes things to be on fire. Who wants a nice big plate of fire?
|
|
#
?
Jul 6, 2022 02:11
|
|
|
|
|
#
?
Jul 6, 2022 03:42
|
|
|
Have a big ol' dump of mostly elfgame bullshit:                      
|
|
#
?
Jul 6, 2022 04:37
|
|
|
ok this is my favorite one yet
|
|
#
?
Jul 6, 2022 12:45
|
|
|
|
|
#
?
Jul 6, 2022 15:03
|
|
|
I really just want to see what it thinks album art should be. For instance: "Sgt. Pepper's Lonely Hearts Club Band" or "The Key of Life"
|
|
#
?
Jul 7, 2022 01:30
|
|
|
A prompt for dall-e 2: “Gods in the blue pyramid, in the style of Enki Bilal”
|
|
#
?
Jul 7, 2022 01:43
|
|
|
down1nit posted:I really just want to see what it thinks album art should be.  
|
|
#
?
Jul 7, 2022 03:30
|
|
|
Popoto posted:A prompt for dall-e 2: 
|
|
#
?
Jul 7, 2022 03:31
|
|
|
lol https://twitter.com/triplux/status/1542988087495499776?t=uWKHkOBkcWw7pWDctCGUQA&s=19
|
|
#
?
Jul 7, 2022 04:17
|
|
|
wa27 posted:Using your prompts: Bottom right: Holy poo poo the AI knows about Billy Preston!
|
|
#
?
Jul 7, 2022 17:27
|
|
|
I’m gonna guess it doesn’t know who Enki Bilal is!
|
|
#
?
Jul 8, 2022 03:39
|
|
|
This one turned out pretty Lewis Carroll-ish:
|
|
#
?
Jul 8, 2022 05:46
|
|
|
has anyone tried "photograph of a hand" or simply "fingers?"
|
|
#
?
Jul 8, 2022 08:38
|
|
|
doctorfrog posted:has anyone tried "photograph of a hand" or simply "fingers?"  and in case you're wondering, no, removing the question mark does not make it less horrifying  sidenote: they've added inline ads to craiyon and the screenshot button now includes them I sincerely hope it's just a mistake, because having to crop the screenshots makes that button pointless TIP fucked around with this message at 09:12 on Jul 8, 2022 |
|
#
?
Jul 8, 2022 09:10
|
|
|
How good is dall-e 2 with foreign terms or concepts? I imagine it doesn't work if you input entire sentences/prompts in another language, but can it figure out what e.g. carabinieri is in context? Something like: "Two carabinieri standing next to their bikes in Tuscany, landscape baroque oil painting" ?
|
|
#
?
Jul 8, 2022 11:30
|
|
|
TIP posted:
See, this is what I've been saying. Try lucid dreaming and then really focusing on your fingers, you'll see horrors like this.
|
|
#
?
Jul 8, 2022 19:51
|
|
|
doctorfrog posted:See, this is what I've been saying. Try lucid dreaming and then really focusing on your fingers, you'll see horrors like this. these AI images really capture the fuckery that goes on in my brain during dreams. the context is there but the specifics are all manner of unreal.
|
|
#
?
Jul 8, 2022 21:07
|
|
|
Deltasquid posted:How good is dall-e 2 with foreign terms or concepts? I imagine it doesn't work if you input entire sentences/prompts in another language, but can it figure out what e.g. carabinieri is in context?  There's no lack of images online with carabinieri though, so it wasn't too complicated. I tried feeding it a similar prompt in Latvian and it just spat out random Latvian-themed images. Substituting carabinieri with Latvian words worked if they were common ones. This was "Two meitenes standing next to their bikes in Tuscany, landscape baroque oil painting" ("meitenes"="girls"): 
|
|
#
?
Jul 8, 2022 21:35
|
|
|
LifeSunDeath posted:these AI images really capture the fuckery that goes on in my brain during dreams. the context is there but the specifics are all manner of unreal. I do think it's pretty interesting that the image AIs seem to be fairly accurately replicating portions of how our brains process imagery the first big AI image thing I remember was deep dream, which managed to capture something about the visual experience of psychedelics that no previous art had managed
|
|
#
?
Jul 8, 2022 21:37
|
|
|
it deffo did multiple symbolic things stacked up on other images in a psychedilic frenzy well enough,
|
|
#
?
Jul 8, 2022 21:55
|
|
|
TIP posted:
Now I’m looking forward to the next version of something like this
|
|
#
?
Jul 9, 2022 01:20
|
|
|
Snowy posted:Now I’m looking forward to the next version of something like this I think the current image AIs are based on the same properties, they look for recognizable patterns in noise and enhance them until they look like something they're just much better at figuring out what makes sense in context instead of turning every image into a kaleidoscope of dogs
|
|
#
?
Jul 9, 2022 01:29
|
|
|
I think the drug dogs were made by fiddling with «neurons» in the model directly, to reveal what a neuron «sees»
|
|
#
?
Jul 9, 2022 01:32
|
|
|
|
| # ? May 30, 2024 13:56 |
|
|
TIP posted:I think the current image AIs are based on the same properties, they look for recognizable patterns in noise and enhance them until they look like something the pixelray site lets you watch it generate. it starts from colorful splotches and transforms over time to an image, it's bonkers. "San Jose Shark forcing a nerd to eat eggs"       https://replicate.com/pixray/text2image
|
|
#
?
Jul 9, 2022 01:35
|
|